humanize-cli

Analyze and humanize AI-generated text. Detect AI markers, score detection risk, and get actionable improvement suggestions.

npm install -g humanize-cli

Why This Exists

AI detectors are everywhere now. GPTZero, Originality.ai, Turnitin — they're flagging content left and right. Problem is, they also flag plenty of human writing.

This tool helps you understand why text gets flagged and how to fix it. Not to help AI pretend to be human, but to help you write in a way that sounds like you actually wrote it.

Quick Start

# Analyze text for AI markers
humanize analyze "This comprehensive solution leverages cutting-edge technology."

# Get detection risk score
humanize score --file article.txt

# Get improvement suggestions
echo "The software utilizes robust methodologies" | humanize suggest

# Auto-transform text
humanize transform "It is important to note that this solution is seamless."

Commands

| Command | Alias | Description | |---------|-------|-------------| | analyze | a | Detect AI markers and patterns | | score | s | Calculate detection risk score | | suggest | sg | Get improvement suggestions | | transform | t | Apply automatic fixes | | watch | w | Watch directory for changes | | config | c | Show configuration and API status |

humanize analyze <text>

Detects AI markers and patterns in your text:

• AI vocabulary (delve, leverage, comprehensive, seamless...)
• Structural patterns (perfect intro/conclusion, uniform sentences)
• Missing human elements (contractions, personal voice, opinions)

$ humanize analyze "This comprehensive guide delves into best practices."

═══ TEXT ANALYSIS ═══

Words: 8 | Sentences: 1 | Avg words/sentence: 8.0

⚠ AI Vocabulary Detected (3)
  • "comprehensive" → try: full, complete, thorough
  • "delves" → try: digs into, explores, looks at
  • "best practices" → try: what works, good approaches

humanize score <text>

Calculates detection risk as a percentage (0-100%):

• 0-20%: Low risk — appears human-written
• 21-40%: Moderate risk — may trigger sensitive detectors
• 41-70%: High risk — likely to be flagged
• 71-100%: Very high risk — will be flagged by most detectors

$ humanize score --file draft.txt

═══ DETECTION RISK SCORE ═══

Overall Risk: 67% (HIGH RISK)

Component Breakdown:
  vocabulary           ████████░░ 78%
  structure            ██████░░░░ 55%
  patterns             ███████░░░ 65%
  naturalness          ██████░░░░ 58%
  personality          ███████░░░ 72%

humanize suggest <text>

Generates prioritized improvement suggestions:

• 🔴 Priority Fixes: Must-fix issues that scream "AI"
• 🟡 Recommended: Improvements that help
• 🟢 Optional Polish: Nice-to-have tweaks
• ⚡ Quick Wins: Copy-paste replacements

$ echo "It is important to note that our solution leverages..." | humanize suggest

═══ IMPROVEMENT SUGGESTIONS ═══

🔴 Priority Fixes
  1. AI vocabulary detected: "leverages"
     → Replace with: use, take advantage of, work with
     Before: "solution leverages cutting-edge..."
     After:  "solution uses cutting-edge..."

  2. No contractions found - text sounds robotic
     → Use contractions! "Do not" → "don't"

humanize transform <text>

Automatically applies humanization fixes:

• Replaces AI vocabulary
• Adds contractions
• Removes filler phrases
• Shows before/after risk score

$ humanize transform "It is important to utilize comprehensive solutions."

═══ TRANSFORMED TEXT ═══

It's important to use complete solutions.

Risk score: 72% → 31% (↓41% improvement)

humanize watch <directory>

Monitor a directory for changes and analyze files in real-time:

$ humanize watch ./content --threshold 60

═══ WATCH MODE ═══

Directory: ./content
Threshold: 60%
Press Ctrl+C to stop

Scanning...

✓ intro.md 23%
⚠ HIGH RISK blog-post.md 78%
  AI words: leverage, comprehensive, robust

Initial scan complete. Analyzed 2 files.
Watching for changes...

Options:

• -t, --threshold <n> — Alert when score >= threshold (default: 70)
• -q, --quiet — Only show high-risk files

humanize config

Check configuration and API key status:

$ humanize config

═══ HUMANIZE CONFIGURATION ═══

API Integration:
  GPTZero API:       ✓ Configured
  Originality.ai:    ✗ Not configured

Detection Settings:
  AI Vocabulary:     90+ words and phrases
  Pattern Detectors: 9 structural patterns
  Scoring:           5-component weighted average

API Integration

Optional integration with GPTZero and Originality.ai for external validation:

# Set API keys
export GPTZERO_API_KEY="your-key-here"
export ORIGINALITY_API_KEY="your-key-here"

# Run with API checks
humanize analyze --api --file article.txt

The tool works fully offline by default. API keys add external validation on top of local analysis.

Input Methods

# Direct text
humanize analyze "Your text here"

# From file
humanize score --file article.txt
humanize score -f draft.md

# Piped input
cat document.txt | humanize suggest
pbpaste | humanize analyze  # macOS clipboard

# Output to file/clipboard
humanize transform -f draft.txt > improved.txt
humanize transform -f draft.txt | pbcopy  # macOS

Options

| Flag | Description | |------|-------------| | -f, --file | Read input from file | | -j, --json | Output results as JSON | | -V, --verbose | Show detailed output | | -q, --quiet | Minimal output | | -a, --api | Include GPTZero/Originality.ai checks | | -t, --threshold | Alert threshold for watch mode (default: 70) | | -h, --help | Show help | | -v, --version | Show version |

What It Detects

AI Vocabulary

Words that scream "I was written by an AI":

• The worst offenders: delve, leverage, utilize, comprehensive, robust, seamless
• Corporate buzzwords: stakeholder, synergy, actionable, paradigm
• Formal transitions: furthermore, moreover, consequently, nevertheless
• AI tells: "It is important to note", "In conclusion", "Great question!"

Structural Patterns

• Perfect intro/body/conclusion structure
• Uniform sentence lengths (low variance)
• Missing contractions
• Overuse of transitions
• Too many bullet points

Missing Human Elements

• No personal pronouns (I, we, you)
• No opinions or perspective
• No emotional language
• No rhetorical questions
• No humor or personality

Documentation

The /docs folder includes comprehensive guides:

• AI Detector Mechanics — How detectors actually work (perplexity, burstiness, entropy)
• Writing Quality Research — What makes writing genuinely good, not just undetectable
• The Humanization Pipeline — Complete 5-stage workflow for transforming AI text
• Tools Setup Guide — Recommended tools and budget breakdowns
• Quick Reference — Cheat sheet for fast fixes

Philosophy

This tool isn't about fooling detectors — it's about writing better.

The same patterns that trigger AI detectors (robotic tone, uniform structure, missing personality) also make writing worse. Whether a human or AI wrote it, text without voice is text nobody wants to read.

Use this tool to:

1. Understand why text sounds robotic
2. Learn what makes writing human
3. Improve your writing habits over time

License

MIT

Built by LXGIC Studios

🔗 GitHub · Twitter